Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Breakfast seminar: The Business Value of Survival Analysis
Evi Nagler
Methodologist - European Renal Best Practice
Renal U...
Agenda
01

Introduction

02

Survival analysis: origin and possible application

03

IBM SPSS modeler and Survival Analysi...
4C Consulting | Our Customers

3
About 4C Consulting | Our service portfolio
Customer Experience Management
Experience Identity | Customer Journeys | Momen...
BI / CI Services

How to
communicate
with them

Customers

How to
optimize your
service to
answer their
needs

Who are the...
Agenda
01

Introduction

02

Survival analysis: origin and possible application

03

IBM SPSS modeler and Survival Analysi...
1

Introductory example
The origin
How to treat?
Metastatic
cancer

Classic
treatment

AFTER 12 MONTHS

New
treatment

30% alive

50% alive

9
How to treat?
Metastatic
cancer

Classic
treatment

AFTER 16 MONTHS

New
treatment

20% alive

21% alive

10
Time is Crucial
Interesting, but how can we use this?
An Application
Customer Churn
How to treat?
Churn

Classic
marketing
program

AFTER 12 MONTHS

New
marketing
program

30% stays

50% stays

15
How to treat?
Churn

Classic
marketing
program

AFTER 16 MONTHS

New
marketing
program

20% stays

21% stays

16
Time is Money
2

The idea
Survival curves | Customer Example
Survival
probability

Event=churn
New marketing program
Classic marketing program

Time...
Survival curves | Traditional Example
Survival
probability

Event=death
New treatment
Classic treatment

Median survival t...
Added value | Entire Sample
=event occurs
=enter the study

Start of study

0

2

4

6

8

10

End of study

12
Time (mont...
Added value | Entire Sample
=event occurs
=enter the study

Start of study

0

2

4

6

8

10

End of study

12
Time (mont...
Added value | Entire Sample
=event occurs
=enter the study

Start of study

0

2

4

6

8

10

End of study

12
Time (mont...
Added value | Entire Sample
=event occurs
=censored

Time in study

An individual censored at time t
should have the same ...
Condition | Non-informative censoring
Hospital A

Hospital B

25
Condition | Non-informative censoring
Hospital A

0

2

4

6

Time (months)

8

Hospital B

10

12

0

2

4

6

8

10

12
...
Condition | Non-informative censoring
Hospital A

Hospital B

27
Condition | Non-informative censoring
Hospital A

0

2

4

6

Time (months)

8

Hospital B

10

12

0

2

4

6

8

10

12
...
Condition | Non-informative censoring

29
Condition | Non-informative censoring
Compare 2 loyalty programs
For who: valuable customers (gold status)

30
Condition | Non-informative censoring
Bank A

Bank B

31
Condition | Non-informative censoring

32
How is your data collected?

Which customers are included?

33
Comparing survival curves
Survival
probability

Treatment A
Treatment B
34
Comparing survival curves
Survival
probability

Event=churn
New marketing program
Classic marketing program

Time (months)...
Randomised trial
Treatment A

All patients

Follow-up

Compare results

RANDOM

Treatment B

Follow-up

36
Observational study
Treatment A

All patients

Follow-up

Compare results

CHOICE

Treatment B

Follow-up

37
Observational study
Campaign A

All patients

Follow-up

Business setting

Compare results

CHOICE

Campaign B

Follow-up
...
3

Modelling

39
Definitions
100%

F(t)=Cumulative Incidence

S(t)=Survival curve
80%

60%

40%

20%

0%
0

1

2

3

4

5

6

7

8

9

10

...
Definitions
30%
Incidence

Hazard

25%
20%
15%
10%
5%
0%
0

1

2

3

4

5

6

7

8

9

10

11

41
Definitions

Time

Cumulative
Survival Curve incidence

Incidence

Hazard

0

100%

0%

20%

20%

1

80%

20%

20%

25%

2...
Definitions

Time

Cumulative
Survival Curve incidence

Incidence

Hazard

0

100%

0%

20%

20%

1

80%

20%

20%

25%

2...
Definitions
30%
Incidence

Hazard

25%
20%
15%
10%
5%
0%
0

1

2

3

4

5

6

7

8

9

10

11

44
Cox proportional hazards model
Most common used model for survival data (*)
Flexible choice of covariates
Fairly easy to m...
Cox proportional hazard model
πœ† 𝑑, 𝒁 = πœ†0 𝑑 𝑒π‘₯𝑝 𝛽1 𝑍1 + 𝛽2 𝑍2 +…+𝛽 𝑝 𝑍 𝑝

πœ†0 =baseline hazard
𝑍1 , 𝑍2 ,… , 𝑍 𝑝 = covariate...
Cox proportional hazard model
πœ† 𝑑, 𝒁 = πœ†0 𝑑 𝑒π‘₯𝑝(𝛽1 𝑍1 )
πœ†0 =baseline hazard
0 = π‘›π‘œ π‘‘π‘Ÿπ‘’π‘Žπ‘‘π‘šπ‘’π‘›π‘‘
𝑍1 =
1 = π‘‘π‘Ÿπ‘’π‘Žπ‘‘π‘šπ‘’π‘›π‘‘

𝛽1 =-0.7 ...
Take home messages
Classic regression ignores time – time is crucial
Solution: survival analysis
Advantages
 Use of entir...
Agenda
01

Introduction

02

Survival analysis: origin and possible application

03

IBM SPSS modeler and Survival Analysi...
Questions?
Let’s have coffee first

50
Upcoming SlideShare
Loading in …5
×

4C Consulting Breakfast Seminar - Survival Analysis

2,273 views

Published on

Published in: Business, Technology
  • Be the first to comment

  • Be the first to like this

4C Consulting Breakfast Seminar - Survival Analysis

  1. 1. Breakfast seminar: The Business Value of Survival Analysis Evi Nagler Methodologist - European Renal Best Practice Renal Unit, Ghent University Hospital Veerle LiΓ©baut Consultant – 4C Consulting Wannes Rosius, Client Technical Professional - IBM SPSS
  2. 2. Agenda 01 Introduction 02 Survival analysis: origin and possible application 03 IBM SPSS modeler and Survival Analysis 04 Closing remarks 05 Q&A
  3. 3. 4C Consulting | Our Customers 3
  4. 4. About 4C Consulting | Our service portfolio Customer Experience Management Experience Identity | Customer Journeys | Moments of Truth | Cross-channel | Unique Customer View | CRM Roadmap | Cultural Change Marketing Excellence β€’ β€’ β€’ β€’ Marketing Maturity Assessment Campaign Management & Automation Campaign Management Outsourcing Marketing Resource Management Sales Excellence β€’ β€’ β€’ β€’ SFA Management & Automation Sales Portfolio Management Sales Middle Office Training & Coaching Service Excellence β€’ Customer Service Automation β€’ Self Service Strategy & Management β€’ Complaints Handling Customer Insight Management Performance Management | Data Quality | Data Mining I Segmentation | Scoring | Profiling | Forecasting | Predictive Analytics 4
  5. 5. BI / CI Services How to communicate with them Customers How to optimize your service to answer their needs Who are they / What do they want DATA 5
  6. 6. Agenda 01 Introduction 02 Survival analysis: origin and possible application 03 IBM SPSS modeler and Survival Analysis 04 Closing remarks 05 Q&A
  7. 7. 1 Introductory example
  8. 8. The origin
  9. 9. How to treat? Metastatic cancer Classic treatment AFTER 12 MONTHS New treatment 30% alive 50% alive 9
  10. 10. How to treat? Metastatic cancer Classic treatment AFTER 16 MONTHS New treatment 20% alive 21% alive 10
  11. 11. Time is Crucial
  12. 12. Interesting, but how can we use this?
  13. 13. An Application
  14. 14. Customer Churn
  15. 15. How to treat? Churn Classic marketing program AFTER 12 MONTHS New marketing program 30% stays 50% stays 15
  16. 16. How to treat? Churn Classic marketing program AFTER 16 MONTHS New marketing program 20% stays 21% stays 16
  17. 17. Time is Money
  18. 18. 2 The idea
  19. 19. Survival curves | Customer Example Survival probability Event=churn New marketing program Classic marketing program Time (months) 19
  20. 20. Survival curves | Traditional Example Survival probability Event=death New treatment Classic treatment Median survival time: 9.6 versus 8 months Time (months) Douillard JY et al. J Clin Oncol 2010; 28 (31): 4697-4705 20
  21. 21. Added value | Entire Sample =event occurs =enter the study Start of study 0 2 4 6 8 10 End of study 12 Time (months) 21
  22. 22. Added value | Entire Sample =event occurs =enter the study Start of study 0 2 4 6 8 10 End of study 12 Time (months) 22
  23. 23. Added value | Entire Sample =event occurs =enter the study Start of study 0 2 4 6 8 10 End of study 12 Time (months) 23
  24. 24. Added value | Entire Sample =event occurs =censored Time in study An individual censored at time t should have the same survival chance as all subject who survive up to time t 0 2 4 6 8 10 12 Time (months) 24
  25. 25. Condition | Non-informative censoring Hospital A Hospital B 25
  26. 26. Condition | Non-informative censoring Hospital A 0 2 4 6 Time (months) 8 Hospital B 10 12 0 2 4 6 8 10 12 Time (months) 26
  27. 27. Condition | Non-informative censoring Hospital A Hospital B 27
  28. 28. Condition | Non-informative censoring Hospital A 0 2 4 6 Time (months) 8 Hospital B 10 12 0 2 4 6 8 10 12 Time (months) 28
  29. 29. Condition | Non-informative censoring 29
  30. 30. Condition | Non-informative censoring Compare 2 loyalty programs For who: valuable customers (gold status) 30
  31. 31. Condition | Non-informative censoring Bank A Bank B 31
  32. 32. Condition | Non-informative censoring 32
  33. 33. How is your data collected? Which customers are included? 33
  34. 34. Comparing survival curves Survival probability Treatment A Treatment B 34
  35. 35. Comparing survival curves Survival probability Event=churn New marketing program Classic marketing program Time (months) 35
  36. 36. Randomised trial Treatment A All patients Follow-up Compare results RANDOM Treatment B Follow-up 36
  37. 37. Observational study Treatment A All patients Follow-up Compare results CHOICE Treatment B Follow-up 37
  38. 38. Observational study Campaign A All patients Follow-up Business setting Compare results CHOICE Campaign B Follow-up We need to adjust for confounders 38
  39. 39. 3 Modelling 39
  40. 40. Definitions 100% F(t)=Cumulative Incidence S(t)=Survival curve 80% 60% 40% 20% 0% 0 1 2 3 4 5 6 7 8 9 10 11 12 Time (months) 40
  41. 41. Definitions 30% Incidence Hazard 25% 20% 15% 10% 5% 0% 0 1 2 3 4 5 6 7 8 9 10 11 41
  42. 42. Definitions Time Cumulative Survival Curve incidence Incidence Hazard 0 100% 0% 20% 20% 1 80% 20% 20% 25% 2 60% 40% 10% 17% 3 50% 50% 42
  43. 43. Definitions Time Cumulative Survival Curve incidence Incidence Hazard 0 100% 0% 20% 20% 1 80% 20% 20% 25% 2 60% 40% 10% 17% 3 50% 50% 43
  44. 44. Definitions 30% Incidence Hazard 25% 20% 15% 10% 5% 0% 0 1 2 3 4 5 6 7 8 9 10 11 44
  45. 45. Cox proportional hazards model Most common used model for survival data (*) Flexible choice of covariates Fairly easy to model Standard software exists Well developed elegant mathematical theory Few distributional assumptions Non informative censoring Proportional hazards Independence (*)Goetghebeur E and Van Rompaye B. Survival analysis edition 2011 45
  46. 46. Cox proportional hazard model πœ† 𝑑, 𝒁 = πœ†0 𝑑 𝑒π‘₯𝑝 𝛽1 𝑍1 + 𝛽2 𝑍2 +…+𝛽 𝑝 𝑍 𝑝 πœ†0 =baseline hazard 𝑍1 , 𝑍2 ,… , 𝑍 𝑝 = covariates 46
  47. 47. Cox proportional hazard model πœ† 𝑑, 𝒁 = πœ†0 𝑑 𝑒π‘₯𝑝(𝛽1 𝑍1 ) πœ†0 =baseline hazard 0 = π‘›π‘œ π‘‘π‘Ÿπ‘’π‘Žπ‘‘π‘šπ‘’π‘›π‘‘ 𝑍1 = 1 = π‘‘π‘Ÿπ‘’π‘Žπ‘‘π‘šπ‘’π‘›π‘‘ 𝛽1 =-0.7 οƒ  exp(𝛽1 )=0.5 47
  48. 48. Take home messages Classic regression ignores time – time is crucial Solution: survival analysis Advantages  Use of entire sample  Instantaneous risk estimation Conditions  Non informative censoring  Proportional hazards  Independence 48
  49. 49. Agenda 01 Introduction 02 Survival analysis: origin and possible application 03 IBM SPSS modeler and Survival Analysis 04 Closing remarks 05 Q&A
  50. 50. Questions? Let’s have coffee first 50

Γ—